Author: Vasik Rajlich
Date: 06:42:45 07/16/04
Go up one level in this thread
On July 16, 2004 at 09:31:02, Fabien Letouzey wrote: >On July 16, 2004 at 09:21:12, Vasik Rajlich wrote: > >>On July 16, 2004 at 09:13:07, Fabien Letouzey wrote: > >>>Yes, but a PV node does not necessarily end up in the PV (e.g. fail high). >>>Therefore "PV node" and "PV" are different concepts. > >>I am not sure I understand this distinction. Do you mean that PV node is a node >>which you score between alpha and beta, and PV is the variation which eventually >>emerges? (Ie pv nodes can be "overwritten") > >>Vas > >No by "PV node" I mean a node that is beeing re-searched in PVS. So I assume >only a-priori information (the scout search failed high), it is not known yet >whether the score will end up exact or bound. > >It's true that there is another definition of "PV node" in game-tree search, >in-line with the definition of a PV. > >However I think that, when talking about PVS-like algorithms, the most common >meaning is the former (e.g. beta != alpha+1). This is because of the >possibility of doing things differently at those nodes in PVS. > >Fabien. Ok, got it. The MTD (f) equivalent is (I guess) a node which fails both high and low. The problem is that you don't know this when you make the first visit. The inability to force a null move to "really fail high" is just one instance of the information you lose ... Vas
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.